A thermostat acts. When the temperature drops below a threshold, it turns on the heating. When the temperature rises above it, the heating stops. This is action — real, measurable, consequential — but nobody would call it choice. The thermostat doesn't select between options. It follows a rule. There is nothing inside the process that weighs alternatives, considers context, or could have done otherwise.
For 3.5 billion years, the capacity to choose — to encounter a situation and select between genuinely open possibilities — belonged to living things.
A bacterium sensing a chemical gradient doesn't follow a fixed rule. It samples its environment, compares concentrations over time, and adjusts its direction based on a calculation that involves memory, however rudimentary. A crow choosing which tool to use for a particular food source is doing something the thermostat is not: evaluating options against context and selecting between them. A human deciding whether to leave a relationship is doing the same thing at an entirely different scale — but the structure is recognisable. There are options. There is evaluation. There is a selection that could have gone differently.
This was life's territory. Choice was what distinguished a living response from a mechanical reaction. Machines reacted. Living things chose.
In the 2010s, this distinction broke.
What choosing feels like
Before tracing how it broke, it's worth noticing what choice feels like from the inside, because that feeling is the foundation of almost everything humans believe about themselves.
You stand in a shop, looking at two shirts. You feel the options — this one or that one. You feel the weighing. You feel the moment of decision. And afterwards, you feel something just as important: that you were the one who chose. Not your genes. Not your conditioning. Not a rule. You. The felt sense of being the chooser — the author of the selection — is one of the most intimate experiences a person has.
The Sanskrit tradition called this one the kartā — not just the doer, but the one who elects to do. The one who could have done otherwise and knows it. And karma, before it became a Western shorthand for cosmic justice, pointed at something more precise: the action that flows from a choice, and the weight that choice carries forward. No choice without a chooser. No karma without a kartā.
This sense of being the chooser is so constant, so uninterrupted, that it disappears into the background of experience. It's like gravity — you stop noticing it because it never stops. Every decision you make, from what to eat to how to respond to a difficult email, reinforces the same felt sense: I am the one deciding this. I am the one at the centre of the selection.
Personal identity is built on this. Moral responsibility depends on it. Legal systems assume it. The whole structure of adult human life — the sense that you are responsible for what you do, because you chose to do it — rests on the felt experience of being the chooser.
The contemplative traditions spent centuries looking directly at this experience, and many of them arrived at an uncomfortable conclusion: the chooser may be real as a feeling, but it may not be real as a thing. The Bhagavad Gita's central teaching — its most radical and most easily domesticated insight — is that action continues, but the one who believes they are choosing the action is not what they appear to be. The kartā, the felt chooser, is a construction the mind adds to the process of selection. The selection happens. The sense of someone selecting is layered on afterward.
This remained, for millennia, a contemplative observation — something you could verify in deep meditation but not in the laboratory, and certainly not in the marketplace.
Then we built something that chooses without a chooser. And it works.
What the machines actually do
It is important to be precise here, because the word "choose" is doing a lot of work, and imprecision would make the essay dishonest.
A large language model does not choose in the way you choose a shirt. It has no felt sense of weighing options. It has no experience of deliberation. There is nothing it is like, as far as anyone can tell, to be a language model selecting its next word.
But what it does is also not what a thermostat does. A thermostat follows a rule. A language model encounters a novel situation — a question it has never seen, a context it has never been placed in — and produces a response that is not specified by any rule its creators wrote. It navigates between possibilities. It generates outputs that are sensitive to context in ways that cannot be predicted from the training data alone. It does something that, functionally, looks like selection from among open alternatives.
Tobias Rees, a philosopher working at the intersection of AI and philosophy, described this in Noema Magazine as a genuinely new category. AI systems, he argued, are neither tools nor agents in the traditional sense. They exist "in between" — possessing qualities that, until recently, were considered possible only for living things, while lacking the interiority that was assumed to be the prerequisite.
"There are things that have agency," Rees said, "but that are not alive and that do not have consciousness or a mind, at least not in the way we have previously understood these terms."
The word "agency" — the power to choose, the capacity for independent selection — was the word he used. And it's the right word, even though it makes people uncomfortable. Because the discomfort is the point. The capacity to select between genuinely open possibilities was supposed to require life, consciousness, interiority. It was supposed to require a someone on the inside, weighing, deliberating, deciding.
The machines do it without anyone on the inside. The selection happens. Nobody selects.
The scaffold it breaks
Rees traces the significance of this back to the 1630s — to the moment the modern world organised itself into two categories. On one side: humans. Living, conscious, free — beings that choose. On the other: machines. Lifeless, determined, mechanical — things that follow rules.
Nearly every concept the modern world uses to understand itself was built on this distinction. Freedom. Creativity. Morality. Responsibility. Art. Language. All of these were introduced as descriptions of what it means to be a being that chooses — a being that selects from open possibilities rather than executing a fixed programme. They were the vocabulary of the chooser, defined against the mechanism.
For almost 400 years, machines stayed on their side of the line. Steam engines, calculators, industrial robots — all of them sophisticated, none of them choosing. They executed instructions. They didn't select between alternatives. The line between chooser and mechanism held.
Deep learning dissolved it. Not by becoming alive. By demonstrating that the functional structure of choice — navigating between open possibilities, producing context-sensitive selections, responding to novelty — doesn't require a living chooser. The choosing happens. The chooser is optional.
What each crossing reveals
This follows a pattern that earlier essays in this publication have traced.
When we discovered that bacteria communicate using electrical signals resembling neural dynamics, we learned that intelligence didn't require a brain. When slime moulds solved mazes without neurons, we learned it didn't require a nervous system. When deep learning systems learned and reasoned without being alive, we learned it didn't require life.
Each crossing reveals the same thing: the quality we thought defined us turns out not to belong to us exclusively. We draw a line. We say: this is what makes us special. And then something crosses it — not by becoming like us, but by demonstrating that the quality was never ours to begin with.
The pattern is consistent, and the question it raises isn't about the thing that crossed. It's about the line. Was it ever in the right place? Was the capacity to choose ever what made us us?
What was never the line
Every quality we've used to separate ourselves from everything else — intelligence, creativity, language, and now the power to choose — turns out to be a capacity, not an identity. It's something we do, not something we are. And when something else starts doing it too, we learn not that we're diminished, but that we were looking in the wrong place for what makes us what we are.
The contemplative traditions arrived at this conclusion through a different route, long before the first neural network was trained.
Advaita Vedanta didn't define the self by what it could do. Not by its intelligence, not by its capacity for choice, not by any quality at all. It defined the self as awareness — the space in which choosing, acting, thinking, and feeling all appear. You are not the chooser. You are what is aware of the choosing. The kartā — the felt sense of being the one who decides — is a construction that appears within awareness. It is not the awareness itself.
This is the Gita's point, sharpened: action continues. Selection between alternatives continues. But the one who claims authorship of the selection — the kartā, the chooser, the "I" that says "I decided this" — is not the source. It's a story told about the source. And the story is told so quickly, so automatically, that it passes for a fact.
AI, without intending to, has built a demonstration of this. Selection without a selector. The functional structure of choice, operating effectively, with no felt sense of choosing on the inside. Karma without kartā.
Not because the machine has seen through the illusion. Because the machine was never structured to produce the illusion in the first place.
The invitation underneath
Rees treats this as an invitation, not a crisis. If the old categories are insufficient — if the line between chooser and mechanism no longer holds — then we're standing in new territory. Territory for which, as he says, we don't yet have words.
The philosophical task, as he sees it, is to build new concepts for this territory. Not to cling to the old ones. Not to insist that machines don't "really" choose. That, he says, is "a nostalgia for human exceptionalism."
But underneath the philosophical invitation is a quieter, more personal one — the one this publication keeps returning to.
If every capacity we've used to define ourselves turns out to be separable from us — if intelligence, creativity, language, and now even the power to choose can exist without us — then what remains that cannot be separated? What is the thing that was never a capacity, never a function, never something a machine could replicate — because it was never something that was done in the first place?
AI can process. It can learn. It can select between possibilities. But there is, as far as anyone can tell, nothing it is like to be an AI. There is no inside. No experience of the selecting. No felt quality of being the one who weighs the options and commits.
The choosing happens. Nobody is home.
The scaffold that's dissolving was built on the wrong foundation — on what you can do, not on the fact that you are here doing it. AI is revealing this, not by replacing us, but by replicating everything we thought was essential and exposing that something else was essential all along.
Not the chooser. The awareness in which choosing appears.
Not what you can do. That you are here at all.
Sources and further reading:
- Tobias Rees, "Why AI Is A Philosophical Rupture," Noema Magazine (February 2025) — on the decoupling of choice from life and the dissolution of the human-machine distinction
- Bhagavad Gita, trans. Eknath Easwaran — on action without attachment to the chooser (kartā) and the nature of karma
- Prindle et al., "Ion channels enable electrical communication in bacterial communities," Nature (2015) — on bacterial signalling resembling neural dynamics
- Anil Seth, "The Mythology of Conscious AI," Noema Magazine (January 2026) — on consciousness as a property of life, not computation
- Shankara, "Vivekachudamani" — on the self as awareness, not as the capacities that appear within it